DNA sequence dataset cleaning remains unresolved

Develop robust, domain-appropriate methodologies for cleaning DNA sequence datasets that address the constraints of the four-symbol nucleotide alphabet and sequencing noise, thereby improving the reliability and accuracy of downstream computational analyses.

Background

The authors note that DNA's four-symbol alphabet creates unique challenges for data preprocessing compared to natural language, where richer symbol sets can aid error detection and correction. They argue that dataset quality and preprocessing significantly limit performance of methylation prediction models.

They propose a linguistic mapping strategy to aid rule-based information extraction and manual cleaning, but they explicitly acknowledge that DNA data cleaning has been a pressing and unresolved issue in bioinformatics.